As also shown in Table 2, the lack of fit (LOF) F-values of all six
models implied the variation of the data around the fitted model
were not significant relative to the pure error. There were 14.08%,
30.75%, 13.13%, 10.43%, 12.55% and 10.25% chance for Y1–Y6 that
the LOF F-value could occur due to noise, respectively. The value of
probability of lack of fit (PLOF) higher than 0.05 show that the Fstatistic
was insignificant, implying significant model correlation
between the variable and process response. Adequate precision
(AP) compared the range of the predicted values at the design
points to the average prediction error. The ratios of six models (15.153–89.894) were greater than 4, which indicated adequate
signals [17]. So these six models could be used to navigate the
design space. Simultaneously, low values of the coefficient of variation
(CV) (1.43–6.05%) indicated good precision and reliability of
the experiments [17]. Besides, the plot of the comparison of actual
and predicted values for overall COD removal (Y1) indicated an adequate
agreement between real data and the ones obtained from the
model (Fig. 1). The other predicted vs. actual value plots for other
five responses (Y2–Y6) were similar to Fig. 1, therefore, they were
not shown in this paper.